• No results found

Epidemic \u3ci\u3eClostridium difficile\u3c/i\u3e Strains Demonstrate Increased Competitive Fitness Compared to Nonepidemic Isolates

N/A
N/A
Protected

Academic year: 2021

Share "Epidemic \u3ci\u3eClostridium difficile\u3c/i\u3e Strains Demonstrate Increased Competitive Fitness Compared to Nonepidemic Isolates"

Copied!
12
0
0

Loading.... (view fulltext now)

Full text

(1)

DigitalCommons@University of Nebraska - Lincoln

DigitalCommons@University of Nebraska - Lincoln

Faculty Publications in Food Science and

Technology

Food Science and Technology Department

2014

Epidemic

Epidemic Clostridium difficile

Clostridium difficile Strains Demonstrate Increased

Strains Demonstrate Increased

Competitive Fitness Compared to Nonepidemic Isolates

Competitive Fitness Compared to Nonepidemic Isolates

Catherine D. Robinson

Michigan State University

Jennifer M. Auchtung

Michigan State University, [email protected]

James Collins

Michigan State University

Robert A. Britton

Michigan State University, [email protected]

Follow this and additional works at:

https://digitalcommons.unl.edu/foodsciefacpub

Part of the

Food Science Commons

Robinson, Catherine D.; Auchtung, Jennifer M.; Collins, James; and Britton, Robert A., "Epidemic

Clostridium difficile Strains Demonstrate Increased Competitive Fitness Compared to Nonepidemic

Isolates" (2014). Faculty Publications in Food Science and Technology. 287.

https://digitalcommons.unl.edu/foodsciefacpub/287

This Article is brought to you for free and open access by the Food Science and Technology Department at DigitalCommons@University of Nebraska - Lincoln. It has been accepted for inclusion in Faculty Publications in Food Science and Technology by an authorized administrator of DigitalCommons@University of Nebraska - Lincoln.

(2)

Competitive Fitness Compared to Nonepidemic Isolates

Catherine D. Robinson, Jennifer M. Auchtung, James Collins, Robert A. Britton

Department of Microbiology and Molecular Genetics, Michigan State University, East Lansing, Michigan, USA

Clostridium difficile infection is the most common cause of severe cases of antibiotic-associated diarrhea (AAD) and is a

signifi-cant health burden. Recent increases in the rate of C. difficile infection have paralleled the emergence of a specific phylogenetic

clade of C. difficile strains (ribotype 027; North American pulsed-field electrophoresis 1 [NAP1]; restriction endonuclease

analy-sis [REA] group BI). Initial reports indicated that ribotype 027 strains were associated with increased morbidity and mortality

and might be hypervirulent. Although subsequent work has raised some doubt as to whether ribotype 027 strains are

hyperviru-lent, the strains are considered epidemic isolates that have caused severe outbreaks across the globe. We hypothesized that one

factor that could lead to the increased prevalence of ribotype 027 strains would be if these strains had increased competitive

fit-ness compared to strains of other ribotypes. We developed a moderate-throughput in vitro model of C. difficile infection and

used it to test competition between four ribotype 027 clinical isolates and clinical isolates of four other ribotypes (001, 002, 014,

and 053). We found that ribotype 027 strains outcompeted the strains of other ribotypes. A similar competitive advantage was

observed when two ribotype pairs were competed in a mouse model of C. difficile infection. Based upon these results, we

con-clude that one possible mechanism through which ribotype 027 strains have caused outbreaks worldwide is their increased

abil-ity to compete in the presence of a complex microbiota.

D

iarrhea and colitis are two of the most common side effects of

antibiotic treatment (

1

).The generally accepted paradigm of

antibiotic-associated diarrhea (AAD) is that antibiotics cause a

perturbation of the intestinal microbiota, presenting conditions

that allow the growth of toxigenic bacteria and viruses. It was not

until the late 1970s that toxigenic Clostridium difficile was

identi-fied as a common causative agent of AAD and colitis (

2

). It is now

estimated that 30% of antibiotic-associated diarrhea cases are

at-tributable to C. difficile (

3

) and that healthy gastrointestinal (GI)

microbial communities play an essential role in providing

coloni-zation resistance to C. difficile infection (

1

,

4

,

5

).

The incidence of C. difficile infection has been steadily rising

over the past decade, and it has recently become the most

com-mon nosocomial infection in the United States (

2

,

6

). This rise in

the rate of C. difficile infection has cooccurred with an increased

prevalence of infection caused by a specific phylogenetic clade of

strains characterized as ribotype 027 (

3

,

7

). Initial clinical and

epidemiological studies reported that ribotype 027 strains were

associated with increased rates of morbidity and mortality,

lead-ing to the hypothesis that these strains are hypervirulent (

7–13

).

However, subsequent analyses comparing the outcomes of

en-demic C. difficile infection caused by ribotype 027 strains to C.

difficile infection caused by strains of other C. difficile ribotypes

have yielded conflicting results. Several studies have found that

infection with ribotype 027 strains did not result in more severe

clinical outcomes across a number of institutions (

14–17

),

whereas other studies have seen higher mortality caused by

ri-botype 027 strains than by strains from some of the other C.

diffi-cile ribotypes (

18

,

19

). What is clear, however, is that ribotype 027

strains have been associated with several large outbreaks; have

undergone rapid, global spread since their emergence; and have

become a prevalent ribotype in many hospitals and regions

(ref-erences

20

and

21

and references therein). Therefore, rather than

ribotype 027 strains being hypervirulent (capable of causing more

severe C. difficile disease), we hypothesized that ribotype 027

strains may instead have increased ecological fitness over strains of

other C. difficile ribotypes.

In order to test this hypothesis, we examined competition

be-tween several ribotype 027 and non-027 strains in human fecal

bioreactors. Fecal bioreactors have previously been used to study

C. difficile invasion of complex microbial communities in vitro, as

well as the effects of potential antibiotic and probiotic treatments

(

4

,

22–24

). When developing our fecal bioreactors, we modified

the parameters of design and operation used in other established

models to allow simpler, higher-throughput fecal minibioreactor

arrays (MBRAs). Using these fecal MBRAs, we examined

compe-tition between four different pairs of ribotype 027 and non-027

clinical isolates. We found that in all competitions studied,

ri-botype 027 strains demonstrated a clear competitive advantage

over non-027 strains, often increasing in abundance more than 2

to 3 orders of magnitude by the end of the experiment. We then

performed similar competitions between ribotype 027 and

non-027 strains in a mouse model of C. difficile infection and saw

sim-ilar increased competitive advantage of the ribotype 027 strains.

These results support our hypothesis that ribotype 027 strains

have become more prevalent due to increased ecological fitness

compared to strains of other C. difficile ribotypes.

Received 27 January 2014 Returned for modification 25 February 2014 Accepted 10 April 2014

Published ahead of print 14 April 2014 Editor: B. A. McCormick

Address correspondence to Robert A. Britton, [email protected]. C.D.R. and J.M.A. contributed equally to this work.

Supplemental material for this article may be found athttp://dx.doi.org/10.1128 /IAI.01524-14.

Copyright © 2014, American Society for Microbiology. All Rights Reserved.

(3)

MATERIALS AND METHODS

MBRA design and operation. MBRAs were designed using computer-assisted design (CAD) software (Argon; Asheller-Vellum) and fabricated with DSM Somos Watershed XC 11122 resin via stereolithography (FineLine prototyping) (Fig. 1). Each MBRA consisted of six reactors, with an internal volume of 25 ml and a working volume of 15 ml. The MBRAs were operated under an atmosphere of 5% CO2-5% H2-90% N2 at 37°C in a heated anaerobic chamber. The media were continuously replenished, and waste was removed at a flow rate of 1.875 ml/h. Prior to use, the MBRAs and media were sterilized by autoclaving and allowed to equilibrate to the anaerobic environment forⱖ72 h. The reactor contents were continuously stirred. Additional details regarding MBRA design and operation are available in the supplemental material.

Strains, media, and growth conditions. All the C. difficile strains used in this study are clinical isolates obtained from the Michigan Department of Community Health (MDCH) (Table 1). They were collected from

Michigan hospitals between December 2007 and May 2008. The MDCH determined the strain toxinotype and North American pulsed-field elec-trophoresis (NAP) status. Ribotyping was determined by Seth Walk (Uni-versity of Michigan). All growth studies were carried out in a 37°C anaer-obic chamber (Coy, Grass Lake, MI) under a 5% CO2-5% H2-90% N2 atmosphere using preequilibrated media.

BHIS and TCCFA were made as previously described (25), except that cysteine was excluded from both media. One liter of bioreactor medium (BRM) contained 1 g tryptone, 2 g proteose peptone number 3, 2 g yeast extract, 0.1 g arabinogalactan, 0.15 g maltose, 0.15 gD-cellobiose, 0.4 g

sodium chloride, 5 mg hemin, 0.01 g magnesium sulfate, 0.01 g calcium chloride, 0.04 g potassium phosphate monobasic, 0.04 g potassium phos-phate dibasic, and 2 ml Tween 80, which were adjusted to pH 6.8 and autoclaved at 121°C for 30 min. Following autoclaving, a filter-sterilized mixture of 1 g taurocholic acid (sodium salt), 40 mgD-glucose, 0.2 g

inulin, 2 g sodium bicarbonate, and 1 mg vitamin K3was added. When needed to solidify media, Bacto agar was added to a final concentration of 1.5% (wt/vol).

Collection and preparation of fecal samples for fecal MBRA experi-ments. Fecal samples were donated by 12 healthy, anonymous donors who were between the ages of 25 and 64, had not taken antibiotics for at least 2 months, and had not consumed probiotic products for at least 2 days prior to donation. Fresh samples were collected in sterile containers, which were then packed in wet ice in a sealed (8.1-quart; Sterilite Ultra-seal) container with two anaerobic GasPaks (BD Biosciences) and trans-ported to the laboratory within 24 h. Samples were then transferred to an anaerobic chamber and manually mixed with sterile equipment. Aliquots were transferred to sterile cryogenic vials and stored at⫺80°C until use. Prior to inoculation, aliquots were pooled in equal masses and resus-pended in sterile, anaerobic phosphate-buffered saline at a concentration of 25% (wt/vol). Pooled samples (pooled at equal volumes) were vortexed vigorously for 5 min, large particulates were removed by centrifugation at 201⫻ g for 5 min, and the supernatants were used for inoculation of the MBRA.

C. difficile invasion and competition growth studies in fecal com-munity MBRAs. BRM was preincubated in the 37°C anaerobic chamber for at least 48 h. Prefilled reactors were then left undisturbed for at least 16 h prior to inoculation; this step was used to validate the sterility of each reactor within the MBRA prior to inoculation with fecal slurry. Four mil-liliters of 25% fecal slurry inoculum, prepared as described above, was inoculated into each reactor with a sterile needle and syringe, giving a final concentration of fecal suspension of⬃5% (wt/vol). Freshly inoculated fecal suspensions were allowed to grow in each reactor within the MBRA for 16 to 18 h prior to the initiation of a flow of fresh medium. After 16 to 18 h, a fresh medium flow and waste removal were initiated; 36 h later, we began dosing clindamycin (final concentration, 500␮g/ml) or an equiv-alent volume of water (solvent for clindamycin) (both stored aerobically at 4°C until use) twice daily for 4 days. Samples (1 ml) were removed from the excess fecal slurry and from the reactors prior to the initiation of dosing and daily thereafter throughout the experiment. The samples were centrifuged at 21,000⫻ g for 1 min, the supernatants were discarded, and

B.

Collect fecal samples from 12

healthy donors Screen for

C. difficile

Aliquot individual samples and store at -80

Pool individual aliquots to make 25% w/v fecal slurry in anaerobic PBS

Vortex fecal slurries for 5 min to homogenize Centrifuge fecal slurries for 5 min at 201 X g to

settle large particulates

Inoculate MBRA with fecal slurry (5% w/v final concentration) Initial cell growth without media flow for 16-18 hrs

Start media flow; continue without perturbation for 36 hrs

Dose with clindamycin twice daily for 4 days

24-48 hr following last clindamycin dose, challenge with vegetative C. difficile cells

Monitor C. difficile proliferation

A.

Sample port Effluent Influent 32.5 mm 200 mm Inner Dimensions: 25 X 25 X 40 mm (25 ml) 47 mm

FIG 1 MBRA design and experimental overview. (A) Example of an MBRA used for cultivation of fecal microbial communities. The placement of the influent, effluent, and sample port for one of the six bioreactor chambers is indicated, as are some of the key dimensions. (B) Flowchart describing the timeline of the bioreactor experiments. PBS, phosphate-buffered saline. Addi-tional details can be found in Materials and Methods.

TABLE 1 Characterization of strains used in this study

Strain Toxinotype

PFGEatype

(NAP status) Ribotype

CD1014 0 MI-NAP4 014 CD2015 III MI-NAP1 027 CD2048 0 MI-NAP3 053 CD3014 0 MI-NAP2 001 CD3017 III MI-NAP1 027 CD4004 0 MI-NAP6 002 CD4010 III MI-UN13 027 CD4015 III MI-NAP1 027

(4)

the cell pellets were stored at⫺80°C until they were subjected to further analysis.

For invasion studies with CD2015, CD2015 was grown in BRM broth batch culture overnight, and the reactors were inoculated either with a 1:100 dilution of the overnight culture (Fig. 2) or from dilutions of the exponentially growing subculture in BRM broth (Fig. 3) (CD2015 con-centrations are specified in the figures) on day 7 of operation. Prior to inoculation, the reactors were tested for C. difficile contamination by se-lective plating of an aliquot of each reactor’s contents on TCCFA supple-mented with rifampin (Rif) (50␮g/ml) and erythromycin (Erm) (20 ␮g/ ml) and by quantitative PCR (qPCR) with C. difficile-specific primers (see below). Additional 200-␮l samples were removed from each reactor either 15 min (Fig. 3) or 3 h (Fig. 2) postinoculation, and CD2015 levels were determined by selective plating on TCCFA with Rif and Erm. On subse-quent days of MBRA operation, C. difficile levels were determined via selective plating of an appropriately diluted 100-␮l aliquot from the 1-ml daily sample, as well as by C. difficile-specific qPCR at the times indicated. For competition studies, bioreactors were set up following the C. dif-ficile in vitro invasion model described above with the following modifi-cations. The strains were inoculated into 5 ml BRM overnight and were subcultured into 10 to 30 ml BRM medium and allowed to grow at 37°C for 4 h before inoculation into the reactors to ensure active growth at the time of inoculation. The subcultures were mixed at various ratios (1:1, 1:2, or 1:5 [027/non-027] as indicated) and inoculated into the reactors. For the replicates indicated in Table S1 in the supplemental material (three replicates of two competition pairs, CD3017-CD3014 and CD4010-CD4004), C. difficile was inoculated into the reactors on day 8 of operation instead of day 7. Samples (0.5 ml) were removed from the reactors imme-diately prior to C. difficile inoculation and 2 h after inoculation and pro-cessed as described above, and 0.25 ml of the aliquots collected prior to C. difficile inoculation was used in qPCRs (described below) to detect possi-ble C. difficile contamination. The MBRAs were run for an additional 10 to 12 days and sampled daily as described above.

Quantitative PCR of the tcdA gene to quantify C. difficile invasion. Frozen culture cell pellets were resuspended in 0.5 ml sterile water and transferred to 2-ml screw-top tubes containing⬃200 ␮l 0.1-mm silica beads (Biospec Products). The samples were homogenized by bead beat-ing (BioSpec Products) on the homogenize settbeat-ing for 1.5 min and cen-trifuged for 1 min at 21,000⫻ g, and the supernatant was transferred to a new tube. When not in use, the processed supernatants were stored at ⫺20°C. C. difficile levels were determined by qPCR with primers specific to the C. difficile toxin A gene (tcdA) (see Table S2 in the supplemental material).

We calibrated the tcdA signal observed in our reactors to a known concentration of C. difficile cells grown under fecal bioreactor conditions in the MBRA and enumerated by plate counting. We processed the sam-ples as described above and spiked them into pooled supernatant samsam-ples prepared from bioreactors prior to C. difficile inoculation. We generated 10-fold dilutions of C. difficile in this background community DNA and used them to generate a standard curve for determining the absolute amounts of C. difficile in our bioreactor samples. We also used 10-fold dilutions of community DNA alone to generate a standard curve for as-sessing the total bacterial signal from our reactors using universal 16S primers (see Table S2 in the supplemental material).

Real-time PCRs were performed in triplicate and contained the fol-lowing components: 4␮l supernatant (undiluted [tcdA] or a 1:500 dilu-tion in sterile water [universal 16S rRNA]), 12.5␮l Power SYBR green PCR master mix (ABI, Carlsbad, CA), 0.25␮l each primer (5 ␮M) (see Table S2 in the supplemental material), and 8␮l Milli-Q water. Real-time PCR was performed using an Eppendorf Mastercycler PCR machine un-der the following conditions: 95°C for 10 min and 40 cycles of 95°C for 15 s followed by 60°C for 1 min. A 20-min melting curve was also performed from 60°C to 95°C. We calculated the tcdA copies per milliliter from our experimental samples using the threshold cycle (CT) values and concen-trations from our standard curve described above. If a sample’s CTvalue fell below the lowest concentration from our standard curve, it was des-ignated below the limit of detection, which was 1,000 tcdA copies/ml. We

B.

A.

7 9 11 13 15 101 102 103 104 105 106 107 Days in Culture C. d iff ic ile CF U/ m l 7 9 11 13 15 101 102 103 104 105 106 107 Days in Culture tc d A c o p ie s/ m l

FIG 2 Fecal bioreactor communities prevent invasion by C. difficile unless disrupted by treatment with clindamycin. We monitored C. difficile prolifera-tion in three independent fecal bioreactors that were either clindamycin treated (reactors 1 to 3; solid squares, circles, and diamonds, respectively) or mock treated (reactors 4 to 6; open squares, circles, and diamonds, respec-tively) through selective plating (A) or quantitative PCR (B). The gray dashed line in panel A represents the theoretical washout rate of nonproliferating C. difficile cells. The dotted line represents the limit of detection for selective plating (A) or qPCR (B) in our experiments.

7 9 11 13 102 103 104 105 106 10 C. difficile CFU/ml

A.

7 Days in Culture

B.

0 1 2 3 4 5 6 10 105 106 107 108 Days in Culture C. difficile CFU/ml 4

FIG 3 C. difficile proliferation was assayed in fecal bioreactors with different levels of inoculum and in pure culture under the continuous-culture condi-tions used for bioreactors. (A) We monitored C. difficile proliferation in four independent fecal bioreactors that were clindamycin treated and inoculated at the indicated densities. (B) We measured C. difficile proliferation in pure cul-ture in three replicate continuous-culcul-ture bioreactors operated under the flow conditions used for fecal bioreactors.

(5)

also determined the total bacterial load per sample based upon the CT value with broad-host-range 16S rRNA primers (see Table S2 in the sup-plemental material). We used these CTvalues, which varied by less than 3 cycles across all samples (CT⫽ 18.75 to 21.97) (see Fig. S1 in the supple-mental material), to normalize the tcdA copy numbers that are reported in Fig. 2.

Preparation of 16S rRNA amplicon sequencing. We extracted DNA from samples using bead beating followed by modified cleanup with a Qiagen DNeasy Tissue Kit as described previously (26). DNA concentra-tions were determined by spectrophotometry at 260 and 280 nm (Nano-drop). We used 40 ng of each DNA as the template in PCR with the following final concentrations of reagents: 200 nM 357F primer, 200 nM 926R primer, 1⫻ AccuPrime PCR Buffer II (Invitrogen), and 0.75 U of AccuPrime Taq DNA High Fidelity (Invitrogen). Primers 357F/962R were designed by the Human Microbiome Project, amplify the V3-V5 variable regions of the 16S rRNA gene, and contain unique barcodes that can be used to multiplex sequencing reactions (27). Each reaction was set up in triplicate and amplified using the following cycle: 95°C for 2 min, followed by 30 cycles of 95°C for 20 s, 50°C for 30 s, and 72°C for 5 min, with a final extension at 72°C for 5 min. Successful PCR amplification products from triplicate reactions were pooled and cleaned with Agencourt AMPure XP (Beckman-Coulter). The products were resuspended with a 0.7⫻ volume of beads, washed twice with 70% ethanol, and eluted with 25␮l of low-EDTA TE buffer (10 mM Tris, 0.1 mM low-EDTA). Concentrations of puri-fied DNA were determined using Quant-It (Invitrogen) according to the manufacturer’s protocol and were pooled in equimolar amounts. Nucle-otide sequencing was performed on a 454 GS Junior (Roche Diagnostics) at Michigan State University (MSU) according to the manufacturer’s pro-tocols.

Processing and analysis of sequencing data. All sequence data were processed using mothur (28) version 1.29.1 (January 2013). Sequences were initially quality filtered using the mothur implementation of PyroNoise to remove low-quality sequences, as well as trimmed and fil-tered to remove those sequences that had any ambiguous bases, mis-matches to the reverse primer or barcode, or homopolymeric stretches longer than 8 nucleotides (nt) and were shorter than 200 nt (29). Se-quences from independent sequencing runs were then compiled into a single fasta file and aligned to the SILVA reference alignment (http://www .arb-silva.de/) using the Nearest Alignment Space Termination (NAST)-based aligner in mothur, trimmed to ensure that sequences overlapped, and preclustered, allowing a difference between sequences of 2 bp or less (29). Potentially chimeric sequences were removed using the mothur im-plementation of UChime (30); the remaining sequences were classified using the Ribosomal Database Project (RDP) training set version 9 (March 2012) and mothur’s implementation of the kmer-based Bayesian classifier. Sequences were binned into operational taxonomic units (OTUs) withⱕ3% sequence dissimilarity using the average neighbor al-gorithm of mothur. Taxonomy was assigned to each OTU based upon the majority sequence consensus within that OTU (31). The number of OTUs, evenness (Simpson evenness), and diversity (inverse Simpson) were calculated using mothur. Differences in OTU abundances between treated and untreated bioreactors were determined using the mothur im-plementation of metastats (32). Bray-Curtis dissimilarities were calcu-lated from the OTU distributions of each sample, which were log10 trans-formed and normalized by dividing the abundance of each OTU in a sample by the maximum abundance observed for that OTU, followed by normalizing the total abundance of OTUs across each sample to the same number using the vegan package in R (33). The metaMDS function of vegan was used to determine the optimal ordination distances for the Bray-Curtis dissimilarities, which were also plotted in R. The significance of community differences was determined by analysis of similarities (ANOSIM), which was calculated in R.

Quantitative-PCR analysis of competition cultures and calculations of the competitive index. The strain-specific genes thyA (027) and thyX (non-027) were used to differentiate strains in competitions. We first

identified the strain specificity of these thymidylate synthase genes while doing in silico genomic comparisons of C. difficile genomes. We then screened a collection of 88 strains belonging to several NAP groups, in-cluding all of the strains used in this study, for the presence of thyA or thyX (see Fig. S3 in the supplemental material) and verified that thyA was unique to NAP1 (ribotype 027) strains by screening. This correlation of

thyA with ribotype 027 strains was noted previously (34).

Frozen culture cell pellets (from 0.5 or 1 ml cells) were washed in the same volume of sterile water, resuspended in the same volume of sterile water, and transferred to 2-ml screw-top tubes containing⬃200 ␮l 0.1-mm silica beads. The samples were then placed in a BeadBeater cell disruptor (BioSpec Products, Bartlesville, OK) on the homogenize setting for 1 min, centrifuged at 21,000⫻ g for 1 min, and diluted 1:10 in sterile water. Real-time PCRs were set up by combining the follow-ing components: 12.5␮l Power SYBR green PCR master mix (ABI, Carlsbad, CA), 0.25␮l each primer (100 ␮M), 11 ␮l Milli-Q water, and 1␮l diluted culture supernatant. The primers used are described in Table S2 in the supplemental material. PCR was performed using the conditions described above. All PCRs were performed in technical triplicate, and the CTvalues are an average of the triplicate data points. The amplification efficiency (E) of each primer set was determined by plotting the CTvalues of a standard curve generated by serial 4-log-unit dilutions of C. difficile template; the sample with the highest signal was diluted in sample with no C. difficile inoculated (community back-ground DNA). Primer efficiencies were calculated using the method de-scribed by Pfaffl: E⫽ 10(⫺1/slope)(35). Competitive indices (CI) were calculated by dividing the endpoint ribotype 027/non-027 ratio by the ratio at time zero (T0) [ratio⫽ 2CT(non-027)⫺ CT(027)]. Primer efficiencies were not factored into the CI calculations; however, they differ by ⬍5% (EthyA⫽ 2.04; EthyX ⫽ 1.95) when calculated from reactions using samples containing C. difficile diluted in fecal community back-ground DNA.

C. difficile competition experiments inhmmice. Germ-free C57/B6

mice were gavaged with fecal slurry pooled from the 12 human fecal do-nors described above. Following initial establishment, a stable humanized microbiota was passed from humanized microbiota mice (hmmice) to their progeny. Descendants of these originalhmmice were maintained under specific-pathogen-free conditions and used for all experiments. To induce susceptibility to C. difficile infection (36), an antibiotic mixture of kanamycin (0.4 mg/ml), gentamicin (0.035 mg/ml), colistin (850 U/ml), metronidazole (0.215 mg/ml), and vancomycin (0.045 mg/ml) was ad-ministered in drinking water ad libitum for 3 days and then replaced with fresh drinking water. After 24 h of plain drinking water, the mice were treated intraperitoneally with clindamycin (10 mg/kg of body weight) and challenged 24 h postinjection with 104pure or mixed 027/non-027 C. difficile spores. The spores were cultivated by spread plating overnight BHIS cultures of C. difficile on BHIS medium and incubating them anaer-obically at 37°C for 5 days. The cells were scraped from the plates, resus-pended in sterile water, and heat treated at 60°C for 30 min to kill vegeta-tive cells, and the viable spores were enumerated by plating appropriate serial dilutions on BHIS supplemented with 0.1% taurocholic acid. Spore preparations were diluted in sterile water to yield the desired concentra-tions (⬃104or⬃105spores/ml) and then mixed, when appropriate, prior to gavaging a total of⬃104spores/mouse. For competition experiments, mice were gavaged with the following ratios of ribotype 027/non-027 spores: 1:14 for CD3017 plus CD1014 and 1:50 for CD4015 plus CD2048. The mice were observed daily for disease symptoms and morbidity. Fecal samples were collected daily and frozen until analyzed.

C. difficile levels were quantified in fecal samples by plating. The fecal samples were weighed, diluted in 500␮l sterile water, and heat treated at 65°C for 30 min to reduce background growth of mouse fecal microbiota. The total number of heat-resistant CFU per gram of feces was determined by spotting appropriate serial dilutions on BHIS plates supplemented with 0.1% taurocholic acid. Ribotype 027-specific heat-resistant CFU per gram of feces were determined by spotting appropriate serial dilutions on

(6)

BHIS plates supplemented with 0.1% taurocholic acid and either 50 ␮g/ml rifampin and 10 ␮g/ml erythromycin (CD4015 plus CD2048 com-petition samples) or 10␮g/ml erythromycin only (CD3017 plus CD1014 competition samples). The plates were incubated anaerobically for 24 to 48 h. Colonies formed on antibiotic-supplemented medium represented levels of ribotype 027 strains (CD3017 and CD4015). Non-027 ribotype strain levels were determined by subtracting the number of colonies formed on selective plates from the number on nonselective plates (total C. difficile). The fecal sample weights were then used to determine the CFU per gram of feces. While TCCFA medium is typically required to adequately reduce background growth and allow accurate enumeration of C. difficile in conventional mouse models of infection, we have found by plating the same fecal samples from our humanized microbiota mice on both TCCFA and BHIS that heat treatment is sufficient to eliminate back-ground growth of the microbiota on BHIS. CI were calculated by dividing the ribotype 027/non-027 strain ratios at day 4 by the ratios in the gavaged spore mixtures. In cases where the number of colonies on selective plates were the same as the number on nonselective plates, the ribotype 027/ non-027 ratio was set to 10. This was determined to be a reasonable level for the non-027 limit of detection based on this type of subtractive anal-ysis, taking into account the plating error, and would most likely result in an underestimate of the 027 CI.

Because we used subtractive plating to measure the levels of ribotype 027/non-027 spores and our detection limit set the maximum observable ribotype 027/non-ribotype 027 strain ratio at 10:1, we found that starting with the ribotype 027 strain in the minority provided a larger dynamic range that allowed us to more readily detect an increase in the ribotype 027 strain levels compared to non-027 levels. Although the ratios of ribotype 027/nonribotype 027 spores gavaged were lower than originally intended, placing the ribotype 027 strains in the minority ensures that the 027 strains do not have an advantage by being at even a slightly higher pro-portion in the mixed spore preparations, since there is always a low level of error in diluting and plating.

RESULTS

Fecal MBRAs provide an in vitro model to study C. difficile

invasion in complex microbial communities. Our objective was

to design human fecal bioreactors that recapitulated

antibiotic-induced C. difficile invasion of a resistant community that also

allowed testing of multiple experimental parameters in replicate

reactors simultaneously. Therefore, we pursued a relatively simple

bioreactor design, an array of six single vessel chambers (MBRA)

with modest operating volume (15 ml) that would allow us to

operate up to 24 continuous-flow fecal bioreactors

simultane-ously in the same anaerobic chamber (

Fig. 1

). The reactors were

fabricated with DSM Somos Watershed XC 11122 resin, which

allowed direct observation of the reactor contents and the ability

to autoclave and reuse the reactors.

Continuous-flow MBRAs were inoculated with fecal samples

pooled from 12 healthy, C. difficile-negative donors, and bacterial

communities were allowed to adapt to growth in culture before

challenging the communities with clindamycin, an antibiotic

known to support C. difficile invasion (

37

). Initial studies

com-pared the abilities of three clindamycin-treated and three

mock-treated communities to resist invasion by C. difficile. We found

that in mock-treated communities challenged with 10

6

vegetative

cells of C. difficile, C. difficile levels were reduced to below the level

of detection in replicate reactors within 1 to 4 days following

in-oculation (

Fig. 2

, open symbols). In contrast, when MBRAs were

treated twice daily with clindamycin for 4 days prior to C. difficile

inoculation, C. difficile levels were maintained at the same high

levels at which they were inoculated for 8 days following

inocula-tion (

Fig. 2

, solid symbols).

Because we used such a high inoculum in this initial

experi-ment, we were interested in establishing the minimum number of

C. difficile cells required for invasion. We found that inoculation

with 10

4

or 150 cells was sufficient to allow clindamycin-induced

invasion by C. difficile and that the final number of CFU per ml

reached 10

5

to 10

6

(

Fig. 3A

). These levels are only

⬃10- to 100-fold

lower than the 10

7

CFU/ml that pure C. difficile reaches in reactors

operating under these continuous-culture conditions (

Fig. 3B

).

In addition to the invasion that we observed for the single

ribotype 027 strain (CD2015) shown in

Fig. 2

, we found similar

invasion dynamics for other ribotype 027 strains (CD3017,

CD4010, and CD4015), as well as strains from other ribotypes

(CD1014, CD3014, and CD4004) (data not shown). Based upon

these results, we concluded that the communities established in

our fecal MBRAs demonstrated the key attribute of an in vitro C.

difficile invasion model that we intended to achieve—the ability to

resist invasion by C. difficile until disrupted by antibiotic

treat-ment.

MBRAs support complex fecal microbial communities. In

order to investigate whether the resistance to invasion that we

observed in our unperturbed MBRA communities was due to the

presence of complex microbial communities or simple

commu-nities composed of a few strains that were inhibitory to C. difficile

growth, we sequenced the V3-V5 hypervariable region of the 16S

rRNA gene from samples collected from the triplicate

clindamy-cin- and mock-treated reactors through pyrosequencing. Samples

were collected prior to antibiotic treatment (day 2) and every 2

days after the initiation of treatment (days 4, 6, 8, 10, and 12; C.

difficile was added to all reactors on day 7), and sequencing data

from these samples were compared to data from duplicate

sam-ples from the initial fecal inoculum.

In order to characterize the community composition of our

samples, we partitioned our sequences into OTUs. These OTUs

are groups of sequences that share

ⱖ97% sequence identity

among their 16S rRNA genes and are separated from sequences in

other OTUs by

ⱖ3% sequence dissimilarity. We chose this level of

sequence identity as the basis of our OTUs because it is often used

as a proxy for species level relationships (

38

), although it has been

acknowledged that this level of sequence similarity does not

al-ways correlate with species level relationships (

39

). At the

se-quence depth examined (1,053 sese-quences/sample), we detected a

mean of 69 OTUs in the untreated (day 2; all reactors) and

mock-treated (days 4, 6, 8, 10, and 12) bioreactor communities (

Fig. 4A

).

The number of OTUs observed in untreated or mock-treated

re-actors was

⬃2.4-fold lower than that observed in the original fecal

inoculum (mean, 168 OTUs) (

Fig. 4A

).

When we compared the composition of the original fecal

inoc-ulum community to that of the bioreactor communities, we found

a significant shift in composition upon culturing in the

bioreac-tors, even by day 2 (

Fig. 4B

). The fecal inoculum was dominated

by members of the phylum Firmicutes, which comprised 74%

3% of the sequences. In contrast, members of the phylum

Bacte-roides were dominant members of the bioreactor communities on

day 2 in culture (

Fig. 4B

), comprising 67%

⫾ 3% of the sequences

in all six replicate reactors studied.

Bioreactor community composition changes in response to

clindamycin treatment. Because clindamycin-treated bioreactor

communities become susceptible to C. difficile invasion (

Fig. 2

),

we anticipated that we would observe changes in the microbial

compositions of these communities compared to the

(7)

mock-treated communities and that these changes would be consistent

with previously published models of C. difficile invasion. One

sig-nificant change we observed in our clindamycin-treated

commu-nities was a significant reduction in the number of OTUs

com-pared to mock-treated reactors (

Fig. 4A

) (P

⬍ 0.01 for days 4 to 12

with two-tailed Student’s t test). The ability of antibiotic

treat-ment to significantly reduce species complexity has been

previ-ously reported (

26

,

40

). Although the species richness declined,

quantitative PCR with broad-range 16S rRNA gene primers

showed the total amount of bacteria in the reactors was equivalent

to that in untreated reactors after clindamycin treatment (see Fig.

S1 in the supplemental material), indicating that C. difficile

inva-sion was not dependent upon a decreased bacterial load in the

bioreactors.

This change in microbial composition was also evident when

comparing the compositions of the communities using the

Bray-Curtis dissimilarity measure, which compares the relative

abun-dances of shared OTUs between communities. When we plotted

these data using nonmetric multidimensional scaling (NMDS)

(

Fig. 5

), we found that all six bioreactor communities were highly

similar prior to treatment on day 2 and that bioreactor

commu-nities diverged in response to clindamycin treatment, as well as

time in culture. ANOSIM (

41

) found strong statistical support

(P

⬍ 0.001) for the distinct partitioning of the communities into

pretreatment, mock treatment, and clindamycin treatment

groups, with clindamycin treatment causing a more significant

shift in community structure than time in culture. Using metastats

(

32

), we identified several specific OTUs that were significantly

different between treated and untreated communities. We

ob-served decreases in specific members of the families

Ruminococ-caceae, Lachnospiraceae, and Clostridiaceae that were consistent

with changes observed in previous animal and human studies (

42

,

43

), providing further support for the relevance of this model for

studying aspects of C. difficile invasion in vitro.

Ribotype 027 strains exhibit a competitive advantage over

non-027 strains in the presence of a complex microbiota.

Hav-ing developed the MBRA C. difficile invasion model, we

investi-gated if ribotype 027 strains were able to better compete than

non-027 ribotype strains for their available niche after antibiotic

treatment in the presence of the complex MBRA communities.

We chose to study recent clinical isolates of ribotype 027 and

non-027 C. difficile strains collected by the Michigan Department of

Community Health in order to avoid confounding effects of strain

adaptation to laboratory conditions. Eighty-eight isolates were

characterized by NAP fingerprint, toxinotype, and ribotype; we

selected eight strains for further study (

Table 1

). Four different

ribotype 027 strains were competed against four different non-027

ribotype strains in order to avoid selecting strains with

unrepre-0 2 4 6 8 10 12 0 50 100 150 200 # of OTU s ( 9 7 % ID ) Days in Culture Firmicutes Bacteriodes Proteobacteria Fusobacteria Synergistetes Unclassified Bacteria Verrucomicrobia Actinobacteria 0.0 0.5 1.0 R e la ti v e A b unda nc e 1 2 1 2 3 4 5 6 Fecal

Inoculum Day 2 Bioreactors

B.

A.

FIG 4 Comparison of community structure between fecal samples and mock-treated and clindamycin-mock-treated reactors. We analyzed the 16S rRNA gene abundances from the three mock-treated and clindamycin-treated communi-ties described in the legend toFig. 2on days 2, 4, 6, 8, 10, and 12 in MBRAs, as well as duplicate samples from the initial fecal inoculum. (A) We plotted the mean number of OTUs in the fecal inoculum (asterisk), and clindamycin-treated (circles) and mock-clindamycin-treated (squares) communities⫾ standard devia-tions. (B) We classified each sequence to the phylum level with at least 80% confidence (sequences at⬍80% were designated unclassified bacteria). We then plotted the relative abundance of each phylum in the duplicate samples of the initial fecal inoculum and the six replicate bioreactor samples from day 2 prior to the initiation of treatment. The reactors are numbered as inFig. 2. (Figure S2 in the supplemental material presents the relative abundance of each phylum in the six replicate bioreactor samples on days 4 to 12.)

−1.5 −1.0 −0.5 0.0 0.5 1.0 1.5 −1.0 −0.5 0.0 0.5 1.0 NMDS Axis 1 NMDS Axis 2 Mock treatment Pre Treatment Clinda. treatment

FIG 5 Community structure changes in response to clindamycin (Clinda.) treatment. We compared community composition between the three replicate clindamycin-treated reactors and three replicate mock-treated reactors de-scribed in the legends toFig. 2and4on days 2, 4, 6, 8, 10, and 12 in culture. Day 2 samples were collected prior to initiation of treatment; samples from days 4 and 6 were collected while clindamycin or mock treatment was ongoing; and samples from days 8, 10, and 12 were collected following the cessation of treatment. C. difficile was added to all six reactors on day 7. We used the Bray-Curtis dissimilarity measure to examine differences in both the presence/ absence and abundance of each OTU across all the samples and plotted these pairwise comparisons using NMDS. Each point on the plot represents a single reactor (clindamycin-treated replicates 1, 2, and 3, solid circles, solid squares, and solid triangles, respectively; mock-treated replicates 1, 2, and 3, open cir-cles, open squares, and open triangles, respectively) at a single point in time. Individual time points for each reactor are connected by lines between the points, beginning on day 2 (pretreatment) and continuing through day 12. From the NMDS, we observed potential partitioning of the samples into three groups (pretreated, mock treated, and clindamycin treated). These distinct distributions were supported by ANOSIM, with P values of⬍0.001. The el-lipses indicate the 95% confidence intervals for the means of the distribution of communities for the indicated groups (pretreatment, mock treatment, and clindamycin treatment). Distinct distributions between clindamycin and ear-ly/mock-treated samples were also supported by ANOSIM, with P values of ⬍0.001. The plot stress for the NMDS was 0.171.

(8)

sentatively high or low competitive fitness for either ribotype

group. For the non-027 ribotypes, we selected strains that were

different ribotypes and had different NAP designations to

broaden the phylogenetic breadth of strains tested.

Exponentially growing pure cultures of ribotype 027 and

non-027 C. difficile strains were mixed together and inoculated into

clindamycin-treated MBRAs. At days 3, 7, and 11 postinoculation,

samples were taken and quantitative PCR was conducted to

deter-mine the relative ratios of the competing strains. Plotted in

Fig. 6

are the CI of ribotype 027 strains at day 7 for all of the replicates in

each competition pair, calculated as the 027/non-027 ratio at day

7 divided by the ratio at day 0. The mean competitive indices

(range) for these competitions are 3.8 (0.75 to 22.4) for CD2015,

91.5 (21.9 to 3,993.2) for CD3017, 19.8 (1.4 to 593.4) for CD4015,

and 152.9 (30.3 to 764.9) for CD4010. Figure S3 in the

supplemen-tal material shows the 027/non-027 ratios of all competition pairs

plotted across time for each individual reactor. The competition

dynamics vary between replicates both within and across

compe-tition pairs. However, there is a strong trend of increasing ratios

over time for the ribotype 027 strains, even when started at

differ-ent initial input ratios, further supporting their competitive

ad-vantage. The competitive indices calculated from these ratios

across all time points (days 3, 7, and 11; some days vary [see the

figure legend]) are reported in Table S1 in the supplemental

ma-terial. Across all 22 competition replicates, only two CI were

⬍1.0

at day 7; they account for two of the six replicates of the CD2015

(ribotype 027) plus CD3014 (ribotype 001) competition pair. This

competition pair was particularly interesting in that the ribotype

027 strain (CD2015) displayed an initial drop in the strain ratio in

the majority of the competition replicates, sometimes emerging as

low as 2% of the total C. difficile population at day 3 (see Fig. S4 in

the supplemental material). Nevertheless, CD2015 was able to

re-cover and eventually outcompete the non-027 ribotype strain by

the end of the competitions. At the day 11 time point of these

replicates, the ratios continued to increase, resulting in CI close to

or

⬎1 (see Table S1 in the supplemental material).

One possible mechanism for ribotype 027 strains to

outcom-pete the non-027 strains would be if the latter were inherently

unable to invade the complex MBRA communities. We tested this

hypothesis by inoculating fecal bioreactors with individual

non-027 strains and saw that the strains were as able to invade the

microbiota as ribotype 027 strains (data not shown). In addition,

the MICs of clindamycin for all of the strains used in the study

ranged from 50

␮g/ml to ⬎100 mg/ml (data not shown),

concen-trations severalfold higher than the calculated residual

clindamy-cin in the reactors at the time of C. difficile inoculation based on

theoretical washout (

⬍9 ␮g/ml). Therefore, the competition

out-come was not reflective of differences in clindamycin sensitivity.

Because we performed our competition experiments in fecal

bioreactors that had been treated with clindamycin and did not

include untreated control reactors, we wanted to verify that the

communities present in our clindamycin-treated competition

re-actors were similar to those previously established and

character-ized in our in vitro invasion model. Therefore, we sequenced the

V3-V5 hypervariable region of the 16S rRNA gene from our

com-petition bioreactor samples at day 7, just prior to C. difficile

inoc-ulation, by pyrosequencing and compared these sequences to the

previous data we had collected from our in vitro invasion model

(

Fig. 4

and

5

). We found that the richness, diversity, and evenness

of the competition communities were similar to those of the other

clindamycin-treated communities on day 6 (see Fig. S5 in the

supplemental material). When we compared the microbial

com-munity structures of the competition bioreactor samples using the

Bray-Curtis dissimilarity measure and plotted them with the

pre-vious clindamycin-treated and mock-treated samples with

NMDS, we found that they grouped together with the

clindamy-cin-treated communities (see Fig. S6 in the supplemental

mate-rial). These community comparisons show that the strains were

competed in the presence of complex, diverse fecal communities

and not community anomalies of unexpectedly low richness or

diversity.

Ribotype 027 strains display a competitive advantage in vivo.

In order to address if ribotype 027 strains are capable of

outpeting non-027 ribotype strains in the intestinal tract, we

com-peted two ribotype 027 and non-027 strains in a humanized

mi-crobiota mouse model of C. difficile infection. Mice were treated

with an antibiotic cocktail (

36

) for 3 days, followed by a single dose

of clindamycin. Twenty-four hours later, the mice were gavaged

with C. difficile spores. Under these conditions, strains CD3017

and CD4015 (ribotype 027) and strains CD1014 and CD2048

(ri-botypes 014 and 053, respectively) were able to transiently

colo-nize the intestinal tracts of the mice when infected individually

without causing severe disease (see Fig. S7 in the supplemental

material).

To compare the relative fitness of ribotype 027 and non-027

strains, antibiotic-treated animals were treated with a mixture of

10

4

spores from strains CD3017 (027) and CD1014 (014) (ratio,

1:14) or CD4015 (027) and CD2048 (053) (ratio, 1:50). The

abun-dance of each strain was monitored daily by selective plating of

mouse feces. The 027 strain competitive indices in replicate mice

for both competition groups are plotted in

Fig. 7

. Competitive

indices were calculated by dividing the 027/non-027 ratios at day 4

by the ratios present in the gavaged spore mixtures.

In both competitions, we noted that the ribotype 027 strains

had a competitive advantage when directly competing in the

mouse intestinal tract (

Fig. 7

). CD3017 displayed a dramatic

ex-FIG 6 Competitive indices of ribotype 027 strains relative to non-027 strains in the presence of MBRA fecal communities. Clindamycin-treated bioreactors were inoculated with the indicated mixtures of strains at various ratios, and the abundance of each strain was monitored over time by qPCR to measure botype 027/non-027 ratios. Plotted here is the competitive index of the ri-botype 027 strains for each replicate competition, calculated as the 027/non-027 ratio at day 7 or 8 (see Table S1 in the supplemental material) divided by the ratio at day 0. Each circle represents an individual replicate competition. Where the non-027 strain was below the limit of detection, the ratio was determined by substituting the highest CTvalue in the linear range (detection limit) for the non-027 value. The horizontal bars represent the geometric means of replicates for each competition.

(9)

pansion over days 2 to 4 when competed against CD1014; at day 4,

the mean CI for 3017 was 65.8 (range, 18.8 to 140.9). CD4015 also

displayed a competitive index that showed it had a competitive

advantage over CD2048, and although the CI was not as robust as

observed with CD3017, it had a competitive advantage, with a CI

of 10.9 (range, 7.9 to 25.0). These data demonstrate that ribotype

027 strains have a competitive advantage over non-027 strains in

vivo.

DISCUSSION

Ribotype 027 strains have been frequently shown to be

overrepre-sented in hospital outbreaks and have been linked to increased

morbidity and mortality. Although this association with a

hyper-virulent state is controversial, the fact that ribotype 027 strains

have swept across the globe implies they have acquired an

in-creased ability to cause disease. We hypothesized that differences

in strain physiology could give ribotype 027 strains a competitive

advantage over strains of other ribotypes, thereby leading to the

increased prevalence of ribotype 027 strains. We used the MBRA

C. difficile invasion model to demonstrate that ribotype 027 strains

were able to outcompete strains of other ribotypes in the presence

of complex fecal bacterial communities. We then demonstrated

competitive advantages of ribotype 027 strains similar to those of

strains of other ribotypes in a mouse model of C. difficile infection.

Our work demonstrates that ribotype 027 strains can directly

outcompete strains of other ribotypes. Because we used four

in-dependent ribotype 027 strains competed against four

indepen-dent strains of various ribotypes, we do not expect that the

ob-served increase in competitive fitness of the ribotype 027 strains

was due to strain selection. In the majority of competition pairs we

studied, the ribotype 027 strains became the dominant C. difficile

strains in the community, sometimes leading to the complete loss

of the non-027 strains. This dominance was observed in all

biore-actor communities studied, as well as in the mouse competition

between the ribotype 027 strain CD3017 and the ribotype 014

strain CD1014. However, in the second mouse competition

be-tween ribotype 027 strain CD4015 and ribotype 053 strain

CD2048, the ratio of CD4015 to CD2048 increased over time,

from a 1:50 ratio at the beginning of the experiment to a mean 1:4

ratio (range, 1:10 to 1:2) at the end of the experiment, but CD4015

did not become the dominant C. difficile strain in the community.

Although we interpret these results to indicate that CD4015 was

outcompeting CD2048 for available niche space and would likely

have led to its loss from the community if the experiment had been

continued further, we cannot exclude the hypothesis that both

strains would have reached a steady level of coexistence. When

comparing CI of competition pairs in the MBRA or mice, similar

trends were observed, with the CD3017 CI higher than the

CD4015 CI in each model, supporting the idea that the MBRA

model can recapitulate the C. difficile dynamics that occur in vivo

and further demonstrating the increased fitness of these strains

and the validity of the MBRA model as a precursor for in vivo

experiments (see Fig. S8 in the supplemental material).

However, it is currently unclear what physiological differences

present in ribotype 027 strains allow these strains to outcompete

strains of other ribotypes. It is unlikely the competitive advantage

of ribotype 027 strains is due to a simple growth rate advantage.

We have performed in vitro batch culture studies in a variety of

different media in our laboratory and found that there were no

significant differences in growth rates of our ribotype 027 strains

compared to strains of other ribotypes (unpublished results).

Fur-ther, recently published data demonstrate that some ribotype 027

strains grow more slowly than strains of other ribotypes (

44

).

Al-though it is possible that ribotype 027 strains are capable of

di-rectly antagonizing strains of other ribotypes, we favor the model

in which ribotype 027 strains have increased cellular fitness that

allows them to better compete for available limiting nutrients,

thereby excluding competing ribotypes and indirectly leading to

their elimination.

One aspect of physiology that could potentially impact

com-petition outcome is interstrain variability in rates of sporulation.

If one strain had a higher proportion of the cells in its population

enter sporulation during the course of competition, that strain

would have fewer cells in vegetative growth, effectively reducing

its competitive fitness. Although there have been some reports

that ribotype 027 strains sporulate more efficiently than other

strains (

45

,

46

), larger studies comparing multiple isolates of

dif-ferent ribotypes have found that there is no significant correlation

between sporulation efficiency and ribotype (

44

,

47

). Because

spo-rulation dynamics of individual strains cocultured within the

MBRA are difficult to measure, we do not currently have data to

determine whether the non-027 strains used in our study

sporu-late to higher levels than the 027 strains. However, when we

as-sayed sporulation in pure culture under both batch and

continu-ous-culture conditions, we did not observe higher rates of

sporulation of these non-027 strains (data not shown). The

dy-namics of C. difficile sporulation in the context of growth in the

presence of a complex fecal community is an area of current and

future investigation. While it is important to consider the impact

of sporulation dynamics on competition outcome, we do not

be-lieve this is the factor responsible for ribotype 027 strains

outcom-peting other ribotypes in our experiments.

A second aspect of physiology that could play a role in

compet-itive fitness is differences in germination. Differential germination

does not play a role in the competitive advantage of ribotype 027

strains in the MBRA, since competitions were initiated with

veg-etative cells. In contrast, competition in the mouse model was

initiated by gavaging a mixture of spores. Recent work published

FIG 7 Competitive indices of ribotype 027 strains relative to non-027 strains in a mouse model of C. difficile infection. After antibiotic treatment, mice were gavaged with mixtures of 027 and non-027 ribotype strain spores. C. difficile abundances for the indicated strains were determined by selective plating of the fecal samples. Plotted here is the competitive index of the 027 strains for each replicate mouse competition as the 027/non-027 ratio at day 4 divided by the ratio at day 0 (ratio in spore mixes). The horizontal bars represent the geometric means of replicates for each competition. The dotted line on the y axis represents the upper CI limit for the CD3017-CD1014 competition based on the plating limit of detection.

(10)

by Francis et al. demonstrates that murine bile acids (muricholic

acids) inhibit C. difficile spore germination and that there is strain

variability in this inhibition (

48

). In addition, published data show

there is significant variability in rates of germination and the

com-pounds that serve as germinants among large sets of C. difficile

strains of varying ribotyes, at least in vitro (

49

). However, when we

compared the day 1 levels of C. difficile in the mice gavaged with

individual strains of different ribotypes to the levels of spores

pres-ent in the initial inoculum, we found similar increases in the

abun-dances of C. difficile cells across the different strains by day 1 (see

Fig. S7 in the supplemental material). Based upon these results, we

conclude that differences in germination rates are unlikely to play

a role in the ribotype 027 competitive advantage.

Comparative genomic studies have identified potential

candi-date genes that could provide ribotype 027 strains an increased

competitive advantage over strains of other ribotypes. One such

genomic difference is the presence of the thymidylate synthase

gene, thyA, in ribotype 027 strains, which has effectively replaced

the native, alternative thymidylate synthase gene, thyX, in the

form of a four-gene insertion (

34

). Escartin et al. showed that

ThyA enzymes have, on average, 10-fold-higher catalytic rates

than ThyX enzymes in vitro and are able to confer higher genome

replication rates in vivo (

50

). Based upon this work, we

hypothe-size that the presence of thyA in ribotype 027 strains may confer a

growth advantage and therefore play a role in competitive fitness.

We have also begun assessing whether differences in carbohydrate

metabolism between ribotype 027 strains and strains of other

ri-botypes could provide a competitive advantage. Preliminary

stud-ies using phenotypic arrays (Biolog) and follow-up growth studstud-ies

have revealed differences in metabolism of some carbohydrates by

ribotype 027 strains (C. D. Robinson and R. Britton, unpublished

results).

In preliminary studies using continuous culture to monitor

competition between pure cultures of ribotype 027 strains and

other ribotypes, we found that ribotype 027 strains also had a

competitive advantage (see Fig. S9 in the supplemental material).

It is difficult to make direct comparisons of the competitive

indi-ces obtained from these pure-culture competitions and the

com-petition experiments in the presence of the complex microbiota

due to differences in flow rates and medium compositions and the

presence of other cells. However, it is clear that the magnitude of

the competitive advantage observed can be variable between the

pure culture and complex-community competitions. Specifically,

the ribotype 027 competition pair with a large competitive index

in both the complex-community bioreactors and mouse

experi-ments (CD3017-CD1014 [014]; mean competitive indices, 91.4 in

community bioreactors and 65.8 in humanized microbiota mice)

showed mixed results (mean competitive index, 0.7) in

pure-cul-ture competitions. In contrast, the competitive indices among the

other ribotype 027 competition pairs tested (CD2015-CD3014

[001] and CD4015-CD2048 [014]) were more similar across the

different types of experiments.

These data suggest that multiple mechanisms likely provide a

competitive advantage to ribotype 027 strains and that

competi-tion in the presence of a complex microbial community may

in-crease the competitive advantage of some ribotype 027 strains.

Colonization resistance by the host microbiota is important in

preventing C. difficile infection and must be overcome in order for

C. diffcile to proliferate in the colon (

4

). Factors mediating

colo-nization resistance include competition for nutrients, antagonism

by production of inhibitory compounds (such as bacteriocins or

short-chain fatty acids), and microbiota-dependent modulation

of host immune functions (reviewed in reference

5

). Based upon

the data that we have presented here, we favor the model in which

competition for nutrients plays an important role in the

compet-itive advantage demonstrated by ribotype 027 strains. However,

we are also investigating whether any of the other factors

mediat-ing colonization resistance could be drivmediat-ing the competition

out-come we observe in our models. One hypothesis that we are

ac-tively investigating is whether ribotype 027 strains might require

less disruption of the microbiota to invade, for example, after

shorter antibiotic regimens or after lower doses of antibiotics,

which might provide a mechanism by which these strains

outcom-pete strains of other ribotypes in order to become so prevalent in

many clinical locations.

Fecal MBRAs as a model for C. difficile invasion. Our goal was

to cultivate complex fecal microbial communities within the

MBRAs that could resist invasion by C. difficile, recognizing that

these communities would not be 1:1 translations of the starting

fecal inoculum. Relative to our initial fecal inoculum, we observed

a significant shift in the ratio of members of the phylum

Bacte-roides to Firmicutes members (

Fig. 4B

). Bacteroides-dominated

communities have been observed in several different in vitro

bio-reactor models (

22

,

51–53

), which often use media of similar

positions for cultivation. Several parameters can affect the

com-positions of the microbial communities that are established in

fecal bioreactors, including the source of fecal material and how it

is processed, the medium composition and turnover time used for

cultivation, and the availability of surfaces for biofilm formation

(reviewed in reference

54

). For example, we chose to pool fecal

samples from 12 donors, reasoning that this might lead to an in

vitro community that was more representative of the microbial

diversity present among different individuals than could be

achieved with a single donor. However, by choosing this pooling

strategy, we may have selected for communities that would not

normally coexist. We have examined the differences between

MBRA communities formed from single and pooled fecal samples

and found that they exhibit similar C. difficile invasion dynamics

(unpublished data). Although modifying different aspects of the

operating parameters could lead to communities with higher

sim-ilarities to the starting fecal inoculum, our results demonstrate

that our current model yields complex fecal MBRA communities

that resist invasion by C. difficile when unperturbed and that are

susceptible when disturbed by antibiotics.

Our MBRA model does not promote invasion of

antibiotic-treated communities by C. difficile spores, and thus, we cannot

monitor spore germination dynamics in the presence of a fecal

microbiota. We are currently attempting to modify the model to

enable spore germination within the MBRAs. Determining those

aspects of the current model inhibitory to spore germination and

outgrowth may also provide new insights into the dynamics

be-tween C. difficile and the microbiota.

In spite of the limitations discussed above, the in vitro model

that we developed allows robust, higher-throughput studies of C.

difficile invasion on shorter time scales than can be accommodated

in animal models and other more complex bioreactor models.

Therefore, this model can serve as a complement to animal studies

by providing a platform for conducting initial,

hypothesis-gener-ating experiments, including experiments regarding potential

therapeutic treatment of C. difficile infection.

References

Related documents